Hi,
maybe I should go a bit more into details.
Hello Barbel.
About your eigenspectrum problem...
Then I followed step-6 to deal with hanging nodes. But this seems
to destroy the stiffness matrix or the eigenvectors. Which means the
eigenvalues no longer converge to the right value.
What is the choice of your error indicator (an eigenvector? a potential?)
and which error estimator are you using (Kelly?). Kelly *should* be ok, if
you get the error indicator right to start with (and start with a
appropriately refined mesh).
We use a residual based error indicator. Therefore we refine the mesh
globally and use SolutionTransfer to have the eigenvectors on the new
mesh. Maybe here already a problem occurs (??). Do I need to distribute
the constraints before I go to the globally refined mesh? Do I need to
compute the constraints on the finer mesh and apply them afterwards?
On the finer mesh we compute a residual using an approximation of the
eigenvalue on the finer mesh by computing a Rayleigh quotient. And here is
what gets really worse. The approximated eigenvalue that I get from the
Rayleigh quotient diverges (really fast). Any ideas?
Thanks a lot in advance,
Bärbel
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
